Trust and trust propagation

Trust is one of the most important properties between people in a social network. Normally trust is a very intangible quality, difficult to define, let alone to quantify, but online social networks give us the opportunity to model it.

We can do this by using simple proxy measures (such as the number of successful transactions on a marketplace network) to give edges between individuals a weighting to show the level of trust between them. This can have important consequences for interaction (for example, Resnick et al (2006) have shown that eBay users with high ratings typically sold at prices over 8% higher than new sellers).

Once we have this kind of trust network we can then use it to guess at trust values between individuals who have not yet interacted, effectively making trust recommendations (I trust you, and you trust her, therefore I should probably trust her too). Guha et al. (2004) talks about a web of trust, where both positive and negative trust experiences propagate through a network.

Trust is also affected by homophily, and the behaviour of communities. In a study of users and product ratings on Epinions and Ciao, Yeung et al (2011) showed that individuals with high trust for each others views tend to become more similar in judgments over time, although they also showed that high trust does not necessarily correlate with similar judgments.

Despite this trust is often used as a means to make recommendations. Golbeck and Hendler (2006) showed how we could use trust as a principle to make movie recommendations. They used a network of movies and reviewers, where reviewers rating of a movie acted a weight on the edge that connected them – essentially the more highly regarded that movie was to that reviewer the stronger the connection. They then used the paths from a reviewer to a movie they had not yet seen (had no connection with) to estimate how much that reviewer would like the film.

In the previous steps we looked at how basic network properties can be used to reveal behavior, and to verify existing social theories. But in the area of trust we see how people are starting to use these properties as building blocks to define higher level ideas.

To what extent do these sorts of trust ratings influence you? For example, the number of likes/favourites on comments, or the average scores given to buyers and sellers on eBay?

Further reading

A summary of Golbeck and Hendler’s film trust paper is attached to this article.

Optionally take a moment to download and read the summary, and in particular to understand their equation for taking several trust paths, and using them to estimate a new trust relationship that is not explicitly defined in the network.

References

  1. Golbeck J and Hendler J. (2006) Filmtrust: movie recommendations from semantic web-based social networks. Consumer Communications and Networking Conference pp. 282-268

  2. Guha, R., Kumar, R., Raghavan, P., & Tomkins, A. (2004). Propagation of trust and distrust (pp. 403–412). Presented at the 13th conference of the World Wide Web, New York, New York, USA: ACM

  3. Resnick, P., Zeckhauser, R., Swanson, J., & Lockwood, K. (2006). The value of reputation on eBay: A controlled experiment. Experimental Economics, 9(2), 79–101.

  4. Yeung, C.-M., & Iwata, T. (2011). Strength of social influence in trust networks in product review sites. Presented at the the fourth ACM international conference on Web Search and Data Mining, New York, New York, USA: (pp. 495–504). ACM Press.

Share this article:

This article is from the free online course:

The Power of Social Media

University of Southampton

Get a taste of this course

Find out what this course is like by previewing some of the course steps before you join: